Le Sun, Aug 25, 2024 at 09:36:26AM +0200, Landry Breuil a écrit :
> Le Sun, Aug 25, 2024 at 03:30:56AM -0400, A Tammy a écrit :
> > 
> > On 8/25/24 3:07 AM, Landry Breuil wrote:
> > > Le Thu, Aug 22, 2024 at 10:04:45AM +0200, Landry Breuil a écrit :
> > >> Le Thu, Aug 22, 2024 at 09:40:23AM +0200, Landry Breuil a écrit :
> > >>> Le Wed, Aug 21, 2024 at 06:11:32PM +0200, Kirill A. Korinsky a écrit :
> > >>>> On Wed, 21 Aug 2024 17:16:11 +0200,
> > >>>> Landry Breuil <landry@openbsd.org> wrote:
> > >>>>> Le Wed, Aug 21, 2024 at 05:00:21PM +0200, Landry Breuil a écrit :
> > >>>>>> Le Wed, Aug 21, 2024 at 03:49:45PM +0200, Kirill A. Korinsky a écrit :
> > >>>>>>> ports@,
> > >>>>>>>
> > >>>>>>> Here a ping from another diff from my
> > >>>>>> can you explain the dance about dropping/copying tests in pre-test ?
> > >>>>>> why was that needed ? isn't doing MODPY_PYTEST_ARGS = --forked enough ?
> > >>>>> now that i've tested, this fails with a rather strange error (strange as
> > >>>>> in "i dont understand why moving tests/ around helps") on all test files:
> > >>>>>
> > >>>> and I confirm that this is the problem I am overlooking by moving tests.
> > >> it feels we're getting close.. but not yet.
> > > this version with a symlink for the rust lib works, and it's probably what im
> > > going to commit. fixing the tests requiring a throwaway synapse test instance
> > > is left for future work.
> > >
> > > Index: Makefile
> > > ===================================================================
> > > RCS file: /cvs/ports/net/synapse/Makefile,v
> > > diff -u -r1.82 Makefile
> > > --- Makefile    31 Jul 2024 16:01:52 -0000      1.82
> > > +++ Makefile    25 Aug 2024 07:04:41 -0000
> > > @@ -72,10 +72,20 @@
> > >  TEST_DEPENDS = ${FULLPKGNAME}:${BUILD_PKGPATH} \
> > >                 devel/py-mock${MODPY_FLAVOR} \
> > >                 devel/py-parameterized${MODPY_FLAVOR} \
> > > +               devel/py-test-forked${MODPY_FLAVOR} \
> > >                 www/py-jwt${MODPY_FLAVOR}
> > >  
> > >  do-configure:
> > >         @${MODCARGO_configure}
> > > +
> > > +MODPY_PYTEST_ARGS =    --forked
> > > +
> > > +# some tests fail, but they need a previously running synapse process, not just a database
> > > +# E synapse.storage.prepare_database.UpgradeDatabaseException: Uninitialised database:
> > > +# run the main synapse process to prepare the database schema before starting worker processes.
> > > +# make sure that the rust library is found
> > > +pre-test:
> > > +       ln -sf ${WRKSRC}/build/lib.openbsd-${OSREV}-${ARCH}-cpython-${MODPY_MAJORMINOR}/synapse/synapse_rust.abi3.so ${WRKSRC}/synapse/
> > >  
> > >  # to generate rust modules.inc:
> > >  #  make modcargo-gen-crates and modcargo-gen-crates-licenses
> > 
> > 
> > I encountered almost exactly the same kind of bug for devel/py-uvloop.
> > The solution was to move or delete the original sources folder, like so
> > -
> > http://cvsweb.openbsd.org/cgi-bin/cvsweb/ports/devel/py-uvloop/Makefile?rev=1.2&content-type=text/x-cvsweb-markup
> > 
> > pre-test:
> > 	cd ${WRKSRC} && test -d uvloop && mv uvloop uvloop-bk || test -d uvloop-bk
> > 
> > This was because python tries to find the package to import in the local
> > folder before it tries to find it in the PYTHONPATH environment vars
> > locations.
> > 
> > Because the python tests run after the fake target has been completed,
> > the move should be fine and shouldn't affect the PLIST generation.
> 
> yeah, i also figured that out after finding how nix runs the tests in
> their synapse package, they rm -Rf synapse before starting tests.
> 
> interestingly, i also realized that upstream doesn't run tests via
> pytest, but via twisted.trial, which might work around the "needs a
> running instance of synapse first" issue.
> 
> currently trying with:
> 
> # this symlink makes sure that the rust library is found, since by default the source is used to run the tests
> pre-test:
>        ln -sf ${WRKSRC}/build/lib.openbsd-${OSREV}-${ARCH}-cpython-${MODPY_MAJORMINOR}/synapse/synapse_rust.abi3.so ${WRKSRC}/synapse/
> 
> do-test:
>        cd ${MODPY_TEST_DIR} && ${SETENV} ${ALL_TEST_ENV} ${MODPY_BIN} -m twisted.trial tests
> 
> this produces a more verbose output with all tests names printed, but maybe
> more tests succeed.

running tests with twisted.trial doesnt require pytest-forked, and only
3 tests fail:

tests.handlers.test_user_directory.UserDirectoryTestCase.test_search_punctuation
twisted.trial.unittest.FailTest: Sequences differ: ['@user-a2:test'] != ['@user-a1:test']

tests.storage.test_room_search.EventSearchInsertionTest.test_null_byte
twisted.trial.unittest.FailTest: 0 != 1

tests.storage.test_room_search.MessageSearchTest.test_sqlite_search
twisted.trial.unittest.FailTest: 0 != 1 : expected 'quick brown' to match 'the quick brown fox jumps over the lazy dog'

FAILED (skips=303, failures=3, successes=3162)

the other skipped tests are for missing deps (or unsupported/not yet
working tests):
- OIDC
- saml2
- pyICU
- jaeger_client
- Postgres
- authlib

so.. i don't know which is best. Keep using pytest --forked as it's more
integrated within the portstree, or run the tests the same way upstream
does ?

Landry